PID-controlled Langevin Dynamics for Faster Sampling of Generative Models
This addresses the efficiency bottleneck for users of Langevin-based generative models, making them more practical for applications where speed is critical, though it is an incremental improvement on existing methods.
The paper tackles the slow sampling speed of Langevin dynamics in generative models by introducing PID-controlled Langevin Dynamics (PIDLD), which uses control-theoretic principles to reduce iterations, achieving higher quality samples with fewer steps in experiments on image generation and reasoning tasks.
Langevin dynamics sampling suffers from extremely low generation speed, fundamentally limited by numerous fine-grained iterations to converge to the target distribution. We introduce PID-controlled Langevin Dynamics (PIDLD), a novel sampling acceleration algorithm that reinterprets the sampling process using control-theoretic principles. By treating energy gradients as feedback signals, PIDLD combines historical gradients (the integral term) and gradient trends (the derivative term) to efficiently traverse energy landscapes and adaptively stabilize, thereby significantly reducing the number of iterations required to produce high-quality samples. Our approach requires no additional training, datasets, or prior information, making it immediately integrable with any Langevin-based method. Extensive experiments across image generation and reasoning tasks demonstrate that PIDLD achieves higher quality with fewer steps, making Langevin-based generative models more practical for efficiency-critical applications. The implementation can be found at \href{https://github.com/tsinghua-fib-lab/PIDLD}{https://github.com/tsinghua-fib-lab/PIDLD}.